Identified a settlement timing error in an online blackjack poker-style side bet, where the payout was incorrectly re-evaluating after player actions (splits) instead of settling at initial deal. Built a combinatorial probability model across 944 scenarios and 9.3 billion card sequences to quantify the mispricing, then executed the strategy over three years with disciplined risk management.
The exploit hinged on a rules-to-code mismatch: the side bet was supposed to settle immediately at deal using only the player's two cards and the dealer's upcard, but the live implementation kept the bet open through splits, letting a later hit card retroactively complete a winning hand.
To quantify the edge, I ran a full combinatorial analysis of all 4-card sequences across a 6-deck shoe (312 × 311 × 310 × 309 = 9,294,695,280 total combinations), calculating P(dealer upcard) × P(your pair) × P(hit card completes hand) for every qualifying scenario, weighted by payout and adjusted for blackjack exclusion factors when the dealer showed a 10 or Ace.
That produced a base return-to-player of 95.38% under standard play, versus 104.24% conditional RTP when the exploit condition was active (roughly an 8.86% chance per hand), for a net player edge of +4.24%. From there I built a Kelly-based staking framework to size bets and survive variance across a high-volume, multi-hand session.